Skip to content

How AI Thinks (Just Enough)

Three Mental Models

You don't need to understand how AI works under the hood. But knowing three things about how it behaves will save you a lot of frustration today — and help you troubleshoot when something feels "off."

1. Probabilistic: Different Every Time

If you asked five different people to summarize the same article, you'd get five different summaries. All accurate, but each one different. AI works the same way.

The same prompt does not always produce the same output. This is called being probabilistic — AI generates responses based on probability, not exact formulas. There are many possible good answers to any question, and each time you ask, AI picks a slightly different path.

This means: - "It worked yesterday" doesn't guarantee it will work the same way today - Two teammates using the same prompt will get different (but both valid) results - Variation is a feature, not a bug — it means you can iterate and refine

You control how much variation you get. Think of every prompt as defining a space of possible answers. A vague prompt like "tell me about avalanche safety" opens a massive space — AI could go in hundreds of directions, so each time you ask, you'll get wildly different results. A specific prompt like "list the five avalanche danger levels with their color codes and recommended actions for backcountry skiers" shrinks that space dramatically — the answers still vary, but they cluster around what you actually want.

This is why the Three Pillars from the previous section work. Scope, Intent, and Structure aren't just formatting tips — they're how you control the variation. Leave them vague and you get surprise. Make them specific and you get useful variety.

2. Stateless: A New Person Every Time

Every conversation with AI starts from scratch. It has no memory of previous conversations. The technical term for this is stateless — it does not retain any information from one session to the next.

Think of it like meeting a new person who has never spoken to you before. Every. Single. Time.

This means: - If you close a chat and start a new one, everything from before is gone - AI won't remember what you built last session or what decisions you made - You need to re-explain context at the start of each new conversation

A note on memory: As of March 2026, some AI chat tools have begun to incorporate cross-conversation memory. These tools can retain certain information between conversations, but they work by saving key facts — not replaying the full chat history. For now, treat each new conversation as a blank slate. If you need AI to know something, tell it directly.

Later in Lift 3, you'll learn how to give AI persistent instructions so it starts every conversation already knowing about your project.

Try It: Ask Twice, Get Two Answers

Try this quick experiment to see the first two concepts in action:

  1. Open your AI chat tool
  2. Ask: "What are the three most important things to know about avalanche safety?"
  3. Read the answer
  4. Start a new conversation (important — don't continue the same one)
  5. Ask the exact same question
  6. Compare the two answers

You should see two different responses — both reasonable, but with different wording, different emphasis, maybe even different items. That's AI being probabilistic. And the fact that starting a new conversation gave you no trace of the first one? That's stateless in action.

Click "New chat" in the sidebar to start a fresh conversation for step 4.

Click "New chat" in the sidebar to start a fresh conversation for step 4.

Quick team check-in: Compare your two answers with a teammate's. Did anyone get the same answer twice? (Spoiler: almost certainly not.)

When variation helps vs. when it hurts: Getting two different answers is actually useful right now — you're exploring, and different perspectives help you discover what matters. But once you know what you want to build and write it down as a user story with acceptance criteria, you want AI to hit the target consistently. That's the shift you'll feel in the workflow: lean into variation during Explore, then constrain it with specificity during Plan and Implement.

3. Context Window: The Oxygen Tank

During a single conversation, AI can only hold so much in its "working memory." This is called the context window, and it's measured in tokens — small chunks of text (roughly 3/4 of a word).

Think of your context window like an oxygen tank. As you talk back and forth, the tank fills up. Eventually, you need to surface for fresh air — start a new conversation so AI can focus clearly again.

Context window as an oxygen tank

When the tank gets full: - AI starts paying less attention to things in the middle of the conversation - Response quality can fade - AI might "forget" things you told it earlier in the same chat

When to start fresh: If you've had 15+ back-and-forth exchanges and AI's responses start feeling repetitive or off-target, that's your signal to start a new conversation.

Key Insight

AI is probabilistic (different every time), stateless (forgets between conversations), and has a limited context window (the oxygen tank). None of these are flaws — they're just how it works. Knowing this means you won't be surprised when AI gives you a different answer today than yesterday, or when it "forgets" something you told it three conversations ago.